[PATCH RFC 11/14] migration/rdma: use multifd to migrate VM for rdma-pin-all mode

[Zhimin Feng]

Signed-off-by: Zhimin Feng <address@hidden>
---
 migration/multifd.c | 15 ++++++++++++
 migration/rdma.c    | 58 +++++++++++++++++++++++++++++++++++++++++----
 migration/rdma.h    |  2 ++
 3 files changed, 70 insertions(+), 5 deletions(-)

diff --git a/migration/multifd.c b/migration/multifd.c
index c986d4c247..ba5e0b11d0 100644
--- a/migration/multifd.c
+++ b/migration/multifd.c
@@ -519,12 +519,27 @@ static void *multifd_rdma_send_thread(void *opaque)
         }
         qemu_mutex_unlock(&p->mutex);
 
+        /* To complete polling(CQE) */
+        while (p->rdma->nb_sent) {
+            ret = qemu_rdma_block_for_wrid(p->rdma, RDMA_WRID_RDMA_WRITE, 
NULL);
+            if (ret < 0) {
+                error_report("multifd RDMA migration: "
+                             "complete polling error!");
+                return NULL;
+            }
+        }
+
         /* Send FINISHED to the destination */
         head.type = RDMA_CONTROL_REGISTER_FINISHED;
         ret = qemu_rdma_exchange_send(p->rdma, &head, NULL, NULL, NULL, NULL);
         if (ret < 0) {
+            error_report("multifd RDMA migration: "
+                         "receiving remote info!");
             return NULL;
         }
+
+        /* sync main thread */
+        qemu_sem_post(&p->sem);
     }
 
 out:
diff --git a/migration/rdma.c b/migration/rdma.c
index 4c48e9832c..873c17dc03 100644
--- a/migration/rdma.c
+++ b/migration/rdma.c
@@ -96,6 +96,23 @@ static const char *wrid_desc[] = {
 
 static const char *rdma_host_port;
 
+/*
+ * index of current RDMA channel for multifd
+ */
+static int current_RDMA_index;
+
+/*
+ * Get the multifd RDMA channel used to send data.
+ */
+static int get_multifd_RDMA_channel(void)
+{
+    int thread_count = migrate_multifd_channels();
+    current_RDMA_index++;
+    current_RDMA_index %= thread_count;
+
+    return current_RDMA_index;
+}
+
 /*
  * Negotiate RDMA capabilities during connection-setup time.
  */
@@ -1328,8 +1345,8 @@ static int qemu_rdma_wait_comp_channel(RDMAContext *rdma)
  * completions only need to be recorded, but do not actually
  * need further processing.
  */
-static int qemu_rdma_block_for_wrid(RDMAContext *rdma, int wrid_requested,
-                                    uint32_t *byte_len)
+int qemu_rdma_block_for_wrid(RDMAContext *rdma, int wrid_requested,
+                             uint32_t *byte_len)
 {
     int num_cq_events = 0, ret = 0;
     struct ibv_cq *cq;
@@ -1731,6 +1748,20 @@ static int qemu_rdma_write_one(QEMUFile *f, RDMAContext 
*rdma,
                                .repeat = 1,
                              };
 
+    /* use multifd to send data */
+    if (migrate_use_multifd() && migrate_use_rdma_pin_all()) {
+        int channel = get_multifd_RDMA_channel();
+        int ret = 0;
+        MultiFDSendParams *multifd_send_param = NULL;
+        ret = get_multifd_send_param(channel, &multifd_send_param);
+        if (ret) {
+            error_report("rdma: error getting multifd_send_param(%d)", 
channel);
+            return -EINVAL;
+        }
+        rdma = multifd_send_param->rdma;
+        block = &(rdma->local_ram_blocks.block[current_index]);
+    }
+
 retry:
     sge.addr = (uintptr_t)(block->local_host_addr +
                             (current_addr - block->offset));
@@ -1948,8 +1979,21 @@ static int qemu_rdma_write_flush(QEMUFile *f, 
RDMAContext *rdma)
     }
 
     if (ret == 0) {
-        rdma->nb_sent++;
-        trace_qemu_rdma_write_flush(rdma->nb_sent);
+        if (migrate_use_multifd() && migrate_use_rdma_pin_all()) {
+            /* The multifd RDMA threads send data */
+            MultiFDSendParams *multifd_send_param = NULL;
+            ret = get_multifd_send_param(current_RDMA_index,
+                                         &multifd_send_param);
+            if (ret) {
+                error_report("rdma: error getting multifd_send_param(%d)",
+                             current_RDMA_index);
+                return ret;
+            }
+            multifd_send_param->rdma->nb_sent++;
+        } else {
+            rdma->nb_sent++;
+            trace_qemu_rdma_write_flush(rdma->nb_sent);
+        }
     }
 
     rdma->current_length = 0;
@@ -3758,7 +3802,10 @@ static int qemu_rdma_registration_stop(QEMUFile *f, void 
*opaque,
     ret = qemu_rdma_exchange_send(rdma, &head, NULL, NULL, NULL, NULL);
 
     if (migrate_use_multifd()) {
-        /* Inform src send_thread to send FINISHED signal */
+        /*
+         * Inform src send_thread to send FINISHED signal.
+         * Wait for multifd RDMA send threads to poll the CQE.
+         */
         int i;
         int thread_count = migrate_multifd_channels();
         MultiFDSendParams *multifd_send_param = NULL;
@@ -3770,6 +3817,7 @@ static int qemu_rdma_registration_stop(QEMUFile *f, void 
*opaque,
             }
 
             qemu_sem_post(&multifd_send_param->sem_sync);
+            qemu_sem_wait(&multifd_send_param->sem);
         }
     }
 
diff --git a/migration/rdma.h b/migration/rdma.h
index 86c89bdd1f..7dc3895698 100644
--- a/migration/rdma.h
+++ b/migration/rdma.h
@@ -281,6 +281,8 @@ int qemu_rdma_exchange_send(RDMAContext *rdma, 
RDMAControlHeader *head,
                             int *resp_idx,
                             int (*callback)(RDMAContext *rdma));
 int qemu_rdma_registration(void *opaque);
+int qemu_rdma_block_for_wrid(RDMAContext *rdma, int wrid_requested,
+                             uint32_t *byte_len);
 
 void rdma_start_outgoing_migration(void *opaque, const char *host_port,
                                    Error **errp);
-- 
2.19.1